In [5]:
plt.imshow(image)
Out[5]:
<matplotlib.image.AxesImage at 0x21c9d52d250>
import numpy as np
import os
import pandas as pd
from matplotlib import pyplot as plt
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from keras.preprocessing import image
import cv2
import random
from tensorflow.keras.optimizers import Adam,SGD
from sklearn.metrics import confusion_matrix, classification_report
data_path = r"C:\Users\ragha\AIML Python\reduced bird image data\train\Common-Kingfisher\Common-Kingfisher_765.jpg"
image = cv2.imread(data_path)
#cv2.imread() method loads an image from the path name
cv2.imread(data_path)
#loads image into 3-D that is for RGB
array([[[145, 174, 195],
[145, 174, 195],
[145, 174, 195],
...,
[150, 181, 204],
[150, 181, 204],
[150, 181, 204]],
[[145, 174, 195],
[145, 174, 195],
[145, 174, 195],
...,
[150, 181, 204],
[150, 181, 204],
[150, 181, 204]],
[[145, 174, 195],
[145, 174, 195],
[145, 174, 195],
...,
[150, 181, 204],
[150, 181, 204],
[150, 181, 204]],
...,
[[155, 184, 205],
[154, 183, 204],
[154, 183, 204],
...,
[142, 173, 198],
[142, 173, 198],
[142, 173, 198]],
[[152, 181, 202],
[153, 182, 203],
[155, 184, 205],
...,
[141, 172, 197],
[141, 172, 197],
[141, 172, 197]],
[[152, 181, 202],
[153, 182, 203],
[155, 184, 205],
...,
[141, 172, 197],
[141, 172, 197],
[141, 172, 197]]], dtype=uint8)
plt.imshow(image)
<matplotlib.image.AxesImage at 0x21c9d52d250>
image.shape
#(height of Image , Wiidth of Image )
(874, 1200, 3)
train = ImageDataGenerator(rescale=1/255)
valid = ImageDataGenerator(rescale=1/255)
BATCH_SIZE = 34
train_dataset = train.flow_from_directory(r'C:\Users\ragha\AIML Python\reduced bird image data\train',
target_size=(256,256),
color_mode='rgb',
class_mode='categorical',
batch_size=BATCH_SIZE,
shuffle=True)
valid_dataset = valid.flow_from_directory(r'C:\Users\ragha\AIML Python\reduced bird image data\valid',
target_size=(256,256),
color_mode='rgb',
class_mode='categorical',
batch_size=BATCH_SIZE,
shuffle=True)
"""
The ImageDataGenerator class is very useful in image classification.
There are several ways to use this generator, depending on the method we use,
here we will focus on flow_from_directory takes a path to the directory containing images sorted in sub directories and
image augmentation parameters.
"""
Found 5100 images belonging to 8 classes. Found 1500 images belonging to 8 classes.
'\nThe ImageDataGenerator class is very useful in image classification. \nThere are several ways to use this generator, depending on the method we use, \nhere we will focus on flow_from_directory takes a path to the directory containing images sorted in sub directories and \nimage augmentation parameters.\n\n'
labels = {value: key for key, value in train_dataset.class_indices.items()}
print("Label Mappings for classes present in the training and validation datasets\n")
for key, value in labels.items():
print(f"{key} : {value}")
Label Mappings for classes present in the training and validation datasets 0 : Asian-Green-Bee-Eater 1 : Brown-Headed-Barbet 2 : Common-Kingfisher 3 : Coppersmith-Barbet 4 : Hoopoe 5 : Indian-Peacock 6 : Indian-Roller 7 : Rufous-Treepie
fig, ax = plt.subplots(nrows=2, ncols=5, figsize=(15, 12))
idx = 0
for i in range(2):
for j in range(5):
label = labels[np.argmax(train_dataset[0][1][idx])]
ax[i, j].set_title(f"{label}")
ax[i, j].imshow(train_dataset[0][0][idx][:, :, :])
ax[i, j].axis("off")
idx += 1
plt.tight_layout()
plt.suptitle("Sample Training Images", fontsize=21)
plt.show()
model = tf.keras.models.Sequential([
tf.keras.layers.Conv2D(128,(3,3), activation = 'relu',input_shape =(256,256,3)),
tf.keras.layers.MaxPool2D(2,2),
tf.keras.layers.BatchNormalization(),
##
tf.keras.layers.Conv2D(64,(3, 3), activation='relu'),
tf.keras.layers.MaxPool2D(2,2),
tf.keras.layers.BatchNormalization(),
##
tf.keras.layers.Conv2D(32,(3, 3), activation='relu'),
tf.keras.layers.MaxPool2D(2,2),
tf.keras.layers.BatchNormalization(),
##
##
tf.keras.layers.Flatten(),
##
tf.keras.layers.Dense(256, activation='relu'),
tf.keras.layers.Dropout(0.5),
tf.keras.layers.Dense(8, activation='softmax')
])
model.summary()
Model: "sequential_2"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_6 (Conv2D) (None, 254, 254, 128) 3584
max_pooling2d_6 (MaxPoolin (None, 127, 127, 128) 0
g2D)
batch_normalization_6 (Bat (None, 127, 127, 128) 512
chNormalization)
conv2d_7 (Conv2D) (None, 125, 125, 64) 73792
max_pooling2d_7 (MaxPoolin (None, 62, 62, 64) 0
g2D)
batch_normalization_7 (Bat (None, 62, 62, 64) 256
chNormalization)
conv2d_8 (Conv2D) (None, 60, 60, 32) 18464
max_pooling2d_8 (MaxPoolin (None, 30, 30, 32) 0
g2D)
batch_normalization_8 (Bat (None, 30, 30, 32) 128
chNormalization)
flatten_2 (Flatten) (None, 28800) 0
dense_4 (Dense) (None, 256) 7373056
dropout_2 (Dropout) (None, 256) 0
dense_5 (Dense) (None, 8) 2056
=================================================================
Total params: 7471848 (28.50 MB)
Trainable params: 7471400 (28.50 MB)
Non-trainable params: 448 (1.75 KB)
_________________________________________________________________
optimizer = Adam(learning_rate=0.001)
model.compile(loss = 'categorical_crossentropy',
optimizer = optimizer,
metrics = ['accuracy'])
model_fit = model.fit(
train_dataset,
epochs = 10 ,
batch_size = 1,
validation_data=valid_dataset)
Epoch 1/10 5100/5100 [==============================] - 852s 167ms/step - loss: 2.0569 - accuracy: 0.1708 - val_loss: 18.8793 - val_accuracy: 0.2420 Epoch 2/10 5100/5100 [==============================] - 845s 166ms/step - loss: 2.0044 - accuracy: 0.2098 - val_loss: 16.5226 - val_accuracy: 0.2780 Epoch 3/10 5100/5100 [==============================] - 851s 167ms/step - loss: 1.9263 - accuracy: 0.2231 - val_loss: 12.7601 - val_accuracy: 0.2773 Epoch 4/10 5100/5100 [==============================] - 864s 169ms/step - loss: 1.8893 - accuracy: 0.2275 - val_loss: 6.5716 - val_accuracy: 0.2847 Epoch 5/10 5100/5100 [==============================] - 869s 170ms/step - loss: 1.8814 - accuracy: 0.2286 - val_loss: 12.3329 - val_accuracy: 0.2753 Epoch 6/10 5100/5100 [==============================] - 877s 172ms/step - loss: 1.8793 - accuracy: 0.2394 - val_loss: 20.5774 - val_accuracy: 0.2793 Epoch 7/10 5100/5100 [==============================] - 866s 170ms/step - loss: 1.8396 - accuracy: 0.2412 - val_loss: 24.0551 - val_accuracy: 0.2680 Epoch 8/10 5100/5100 [==============================] - 862s 169ms/step - loss: 1.8505 - accuracy: 0.2402 - val_loss: 9.6119 - val_accuracy: 0.2847 Epoch 9/10 5100/5100 [==============================] - 863s 169ms/step - loss: 1.8082 - accuracy: 0.2455 - val_loss: 13.7061 - val_accuracy: 0.2827 Epoch 10/10 5100/5100 [==============================] - 862s 169ms/step - loss: 1.8866 - accuracy: 0.2427 - val_loss: 20.3183 - val_accuracy: 0.2807
model_fit.history
{'loss': [2.0568881034851074,
2.004366874694824,
1.9262986183166504,
1.889298439025879,
1.8813776969909668,
1.879347801208496,
1.8395916223526,
1.8505151271820068,
1.8082141876220703,
1.886619210243225],
'accuracy': [0.17078430950641632,
0.20980392396450043,
0.2231372594833374,
0.22745098173618317,
0.22862745821475983,
0.23941177129745483,
0.24117647111415863,
0.2401960790157318,
0.2454901933670044,
0.24274510145187378],
'val_loss': [18.87932777404785,
16.52264404296875,
12.760092735290527,
6.571611404418945,
12.332890510559082,
20.577449798583984,
24.05507469177246,
9.6119384765625,
13.70605754852295,
20.318269729614258],
'val_accuracy': [0.24199999868869781,
0.27799999713897705,
0.2773333191871643,
0.28466665744781494,
0.27533334493637085,
0.27933332324028015,
0.2680000066757202,
0.28466665744781494,
0.2826666533946991,
0.28066667914390564]}
def acc_curve(history):
plt.plot(history.history['accuracy'], label='train_acc')
plt.plot(history.history['val_accuracy'], label='val_acc')
plt.title('Training and Validation Accuracy', fontsize=20)
plt.xlabel('Epochs', fontsize = 15)
plt.ylabel('Accuracy', fontsize = 15)
plt.legend()
plt.show()
acc_curve(model_fit)
def loss_curve(history):
plt.plot(history.history['loss'], label='train_loss')
plt.plot(history.history['val_loss'], label='val_loss')
plt.title('Training and Validation Loss', fontsize=20)
plt.xlabel('Epochs', fontsize=15)
plt.ylabel('Loss', fontsize=15)
plt.legend()
plt.show()
loss_curve(model_fit)
model2 = tf.keras.models.Sequential([
tf.keras.layers.Conv2D(128,(3,3), activation = 'relu',input_shape =(256,256,3)),
tf.keras.layers.MaxPool2D(2,2),
tf.keras.layers.BatchNormalization(),
##
tf.keras.layers.Conv2D(64,(3, 3), activation='relu'),
tf.keras.layers.MaxPool2D(2,2),
tf.keras.layers.BatchNormalization(),
##
tf.keras.layers.Conv2D(32,(3, 3), activation='relu'),
tf.keras.layers.MaxPool2D(2,2),
tf.keras.layers.BatchNormalization(),
##
##
tf.keras.layers.Flatten(),
##
tf.keras.layers.Dense(256, activation='relu'),
tf.keras.layers.Dropout(0.5),
tf.keras.layers.Dense(8, activation='softmax')
])
model2.summary()
Model: "sequential_5"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_15 (Conv2D) (None, 254, 254, 128) 3584
max_pooling2d_15 (MaxPooli (None, 127, 127, 128) 0
ng2D)
batch_normalization_15 (Ba (None, 127, 127, 128) 512
tchNormalization)
conv2d_16 (Conv2D) (None, 125, 125, 64) 73792
max_pooling2d_16 (MaxPooli (None, 62, 62, 64) 0
ng2D)
batch_normalization_16 (Ba (None, 62, 62, 64) 256
tchNormalization)
conv2d_17 (Conv2D) (None, 60, 60, 32) 18464
max_pooling2d_17 (MaxPooli (None, 30, 30, 32) 0
ng2D)
batch_normalization_17 (Ba (None, 30, 30, 32) 128
tchNormalization)
flatten_5 (Flatten) (None, 28800) 0
dense_10 (Dense) (None, 256) 7373056
dropout_5 (Dropout) (None, 256) 0
dense_11 (Dense) (None, 8) 2056
=================================================================
Total params: 7471848 (28.50 MB)
Trainable params: 7471400 (28.50 MB)
Non-trainable params: 448 (1.75 KB)
_________________________________________________________________
optimizer = Adam(learning_rate=0.001)
model2.compile(loss = 'categorical_crossentropy',
optimizer = optimizer,
metrics = ['accuracy'])
model2_fit = model2.fit(
train_dataset,
epochs = 10,
batch_size = 34 ,
validation_data=valid_dataset)
Epoch 1/10 150/150 [==============================] - 370s 2s/step - loss: 7.5438 - accuracy: 0.3704 - val_loss: 6.1856 - val_accuracy: 0.1053 Epoch 2/10 150/150 [==============================] - 369s 2s/step - loss: 2.6936 - accuracy: 0.5280 - val_loss: 9.7825 - val_accuracy: 0.1973 Epoch 3/10 150/150 [==============================] - 368s 2s/step - loss: 1.2989 - accuracy: 0.6267 - val_loss: 1.8677 - val_accuracy: 0.3433 Epoch 4/10 150/150 [==============================] - 369s 2s/step - loss: 0.8165 - accuracy: 0.7359 - val_loss: 1.6311 - val_accuracy: 0.5320 Epoch 5/10 150/150 [==============================] - 369s 2s/step - loss: 0.6691 - accuracy: 0.7802 - val_loss: 1.7864 - val_accuracy: 0.5900 Epoch 6/10 150/150 [==============================] - 367s 2s/step - loss: 0.4908 - accuracy: 0.8341 - val_loss: 1.7801 - val_accuracy: 0.6213 Epoch 7/10 150/150 [==============================] - 367s 2s/step - loss: 0.4846 - accuracy: 0.8412 - val_loss: 2.2733 - val_accuracy: 0.5367 Epoch 8/10 150/150 [==============================] - 371s 2s/step - loss: 0.3757 - accuracy: 0.8749 - val_loss: 1.9912 - val_accuracy: 0.6400 Epoch 9/10 150/150 [==============================] - 368s 2s/step - loss: 0.3403 - accuracy: 0.8890 - val_loss: 1.7403 - val_accuracy: 0.6513 Epoch 10/10 150/150 [==============================] - 367s 2s/step - loss: 0.3025 - accuracy: 0.9004 - val_loss: 2.0036 - val_accuracy: 0.6287
model2_fit.history
{'loss': [7.543827533721924,
2.6935853958129883,
1.2989445924758911,
0.8164703845977783,
0.6691080927848816,
0.4908428192138672,
0.4846170246601105,
0.3756808936595917,
0.3403386175632477,
0.30252203345298767],
'accuracy': [0.3703921437263489,
0.5280392169952393,
0.6266666650772095,
0.7358823418617249,
0.7801960706710815,
0.8341176509857178,
0.841176450252533,
0.8749019503593445,
0.8890196084976196,
0.9003921747207642],
'val_loss': [6.1856231689453125,
9.782462120056152,
1.8676767349243164,
1.6310862302780151,
1.786400556564331,
1.7800902128219604,
2.273264169692993,
1.9912046194076538,
1.7402656078338623,
2.0036211013793945],
'val_accuracy': [0.10533333569765091,
0.19733333587646484,
0.34333333373069763,
0.5320000052452087,
0.5899999737739563,
0.6213333606719971,
0.5366666913032532,
0.6399999856948853,
0.6513333320617676,
0.6286666393280029]}
acc_curve(model2_fit)
loss_curve(model2_fit)
test = r'C:\Users\ragha\AIML Python\reduced bird image data\valid'
test_datagen = ImageDataGenerator(rescale=1.0/255)
test_dataset = test_datagen.flow_from_directory(test,
target_size=(256,256),
color_mode='rgb',
class_mode='categorical',
batch_size=50,
shuffle=True)
Found 1500 images belonging to 8 classes.
predictions = model2.predict(test_dataset)
30/30 [==============================] - 22s 737ms/step
fig, ax = plt.subplots(nrows=2, ncols=5, figsize=(12, 10))
idx = 0
for i in range(2):
for j in range(5):
predicted_label = labels[np.argmax(predictions[idx])]
ax[i, j].set_title(f"{predicted_label}")
ax[i, j].imshow(test_dataset[0][0][idx])
ax[i, j].axis("off")
idx += 1
plt.tight_layout()
plt.suptitle("Test Dataset Predictions", fontsize=20)
plt.show()
test_loss, test_accuracy = model2.evaluate(test_dataset , batch_size = 200)
30/30 [==============================] - 23s 754ms/step - loss: 2.0036 - accuracy: 0.6287
print(f"Test Loss: {test_loss}")
print(f"Test Accuracy: {test_accuracy}")
Test Loss: 2.0036208629608154 Test Accuracy: 0.6286666393280029
y_pred = np.argmax(predictions, axis=1)
y_true = test_dataset.classes
import seaborn as sns
cf_mtx = confusion_matrix(y_true, y_pred)
group_counts = ["{0:0.0f}".format(value) for value in cf_mtx.flatten()]
group_percentages = ["{0:.2%}".format(value) for value in cf_mtx.flatten()/np.sum(cf_mtx)]
box_labels = [f"{v1}\n({v2})" for v1, v2 in zip(group_counts, group_percentages)]
box_labels = np.asarray(box_labels).reshape(8, 8)
plt.figure(figsize = (12, 10))
sns.heatmap(cf_mtx, xticklabels=labels.values(), yticklabels=labels.values(),
cmap="YlGnBu", fmt="", annot=box_labels)
plt.xlabel('Predicted Classes')
plt.ylabel('True Classes')
plt.show()
errors = (y_true - y_pred != 0)
y_true_errors = y_true[errors]
y_pred_errors = y_pred[errors]
test_images = test_dataset.filenames
test_img = np.asarray(test_images)[errors]
fig, ax = plt.subplots(nrows=2, ncols=5, figsize=(12, 10))
idx = 0
for i in range(2):
for j in range(5):
idx = np.random.randint(0, len(test_img))
true_index = y_true_errors[idx]
true_label = labels[true_index]
predicted_index = y_pred_errors[idx]
predicted_label = labels[predicted_index]
ax[i, j].set_title(f"True Label: {true_label} \n Predicted Label: {predicted_label}")
img_path = os.path.join(test, test_img[idx])
img = cv2.imread(img_path)
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
ax[i, j].imshow(img)
ax[i, j].axis("off")
plt.tight_layout()
plt.suptitle('Wrong Predictions made on test set', fontsize=20)
plt.show()
